BP Neural Network Optimization Model Based on Nonlinear Function Transformation Approach

Regular harmony search algorithm has defects such as prematureness and convergence stagnation when treating complicated optimization problems, which influence on optimizing the performance of BP neural network. For function optimization problems, we analyze two key parameters of HS algorithm: harmony fine adjustment probability and the harmony adjustment range, which affect the performance in searching. Then we propose a dynamic method based on the adaptive change of PAR and BW. The improved HS algorithm is integrated with BP neural network to optimize the network weight. The simulation results show that in the optimization, the algorithm proposed in this paper is better than basic HS and other improved HS algorithms. IT reduce the network error obviously and speeds up the convergence rate.